This paper describes the use of simulations to model the realized outlet sample for the commodities and services component of the U.S. Consumer Price Index as part of the effort to optimize the sample design. Outlet samples for different categories are drawn from independent frames obtained through the Telephone Point of Purchase survey. Outlets may occur in multiple categories or half samples, so the realized sample may have fewer unique outlets than the initial sample size specified. The efforts to model the actual number of unique outlets realized in spite of difficulties arising from the use of several sources of data are documented, and an attempt is made to evaluate the effectiveness of the resulting models. As the sample composition is changing over time from one drawn entirely from frames obtained through the Consumer Point of Purchase Survey to one drawn entirely from frames obtained through the Telephone Point of Purchase Survey an attempt is made to assess how the introduction of frames from this new source affects the realized sample.